Bizzare as.POSIXct behavior for some time stamp with time part dropping

Question

I was struggling that read.csv with colClasses containing POSIXct was rounding up entire timestamps column down to date dropping time part. I came across a similar question suggesting that some date I have may miss time part. That was not the case. However, after bisecting my vector, I noticed that some particular time stamps are to blame. Here is a snippet.

as.POSIXct(c("2016-03-13 01:00:00", "2016-03-13 02:00:00", "2016-03-13 03:00:00"))

That yields to me

[1] "2016-03-13 CST" "2016-03-13 CST" "2016-03-13 CST"

It is around DST transition, but nevertheless where is time part? Is it a bug?

> version
               _                           
platform       i386-w64-mingw32            
arch           i386                        
os             mingw32                     
system         i386, mingw32               
status                                     
major          3                           
minor          3.0                         
year           2016                        
month          05                          
day            03                          
svn rev        70573                       
language       R                           
version.string R version 3.3.0 (2016-05-03)
nickname       Supposedly Educational      

Update

While setting time zone globally, seems to overcome the problem, it still looks like a bug to me.

Update 2

I confirm that behavior is Windows specific (platform specific bug?) here is the output from R 3.2.3 on Ubuntu

[1] "2016-03-13 01:00:00 CST" "2016-03-13 01:00:00 CST"
[3] "2016-03-13 03:00:00 CDT"

Update 3

There is a known unconfirmed bug #16852.

Update 4

Unless I'm missing something, there is no difference in my case between %S and %OS as mentioned in the comment.

> strptime(c("2016-03-13 01:00:00", "2016-03-13 02:00:00", "2016-03-13 03:00:00"), "%Y-%m-%d %H:%M:%S")
[1] "2016-03-13 01:00:00 CST" "2016-03-13 02:00:00"     "2016-03-13 03:00:00 CDT"
> strptime(c("2016-03-13 01:00:00", "2016-03-13 02:00:00", "2016-03-13 03:00:00"), "%Y-%m-%d %H:%M:%OS")
[1] "2016-03-13 01:00:00 CST" "2016-03-13 02:00:00"     "2016-03-13 03:00:00 CDT"

P.S. I didn't dig into the code yet... :/

Aurelea

I doubt it's a bug. Due to DST, the datetime "2016-03-13 02:00:00" doesn't actually exist. It literally never happened. So as.POSIXct is probably just trying to be consistent in what it returns across the vector. Note that changing just that element to 01:59:59 works. – Terrapin

@Terrapin I wouldn't complain if it just gave me an error, or turned that into 01:00:00 with whatever time zone. But to silently drop time part...., took me quite some time to pinpoint the issue. – Aurelea

Well, I'm also fairly certain that most of these low-level datetime functions in R are simply using system level tools specific to your OS. So you may have Windows to blame, not R. – Terrapin

As I'm digging through as.POSIXlt.character I'm beginning to reconsider my initial reaction to this... – Terrapin

...specifically, compare the effect of strptime on your vector using a format of either "%Y-%m-%d %H:%M:%S" or "%Y-%m-%d %H:%M:%OS", the latter of which is what is eventually hit in as.POSIXlt.character, called from as.POSIXct.default. The %OS format is documented at the end of the Details section of ?strptime. – Terrapin

Interesting. On my platform (OS X) it works with %S but with %OS is removes the times completely. – Terrapin

On Windows 7, I can see that lapply(c("2016-03-13 01:00:00", "2016-03-13 02:00:00", "2016-03-13 03:00:00"), as.POSIXct, tz="America/Chicago") doesn't strip the times from the 1st and 3rd entries. Strange that it would give a different result when treated separately instead of together. – Endgame

The R bug #16852 started via discussion on data.table issue 1619. – Wraf